• Schedule: Friday, December 12, 12:45-1:45
    Location: Asia I+II+III+IV

    A quest to tame complex control systems: from theory towards applications

    Continuous innovation developments have resulted and still are resulting in a dominating trend towards increasing complexity of control systems. The systems of systems that are required for example for the energy transition require novel ways to handle the increasing complexity. In this Bode lecture, a brief overview of the theory about balanced realizations that is based on input-output structure and the model reduction that can be done based on this will be provided. Then, other relevant structure will be discussed, such as dissipativity, multi-physics structure from the port-Hamiltonian, Brayton-Moser, and Euler-Lagrange frameworks, network structure, uncertainty structure, structure in data, etc. Computational aspects, the relation with abstraction and uncertainty and error analysis will be discussed as well. Furthermore, the recent increase of available data and its impact on the methods will be discussed as well. Illustrations of the theory for large scale energy systems and for various high-tech systems will be provided.  The presentation highlights joint work with many collaborators, including students and post-docs. 

  • Schedule: Wednesday, December 10 | 8:00-9:00
    Location: Asia I+II+III+IV

    Using transient controllers to satisfy high level multi-robot tasks

    Multi-robot task planning and control under temporal logic specifications has been gaining increasing attention in recent years due to its applicability among others in autonomous systems, manufacturing systems, service robotics and intelligent transportation. Initial approaches considered qualitative logics, such as Linear Temporal Logic, whose automata representation facilitates the direct use of model checking tools for correct-by-design control synthesis. In many real world applications however, there is a need to quantify spatial and temporal constraints, e.g., in order to include deadlines and separation assurance bounds. This led to the use of quantitative logics, such as Metric Interval and Signal Temporal Logic, to impose such spatiotemporal constraints. However, the lack of direct automata representations for such specifications hinders the use of standard verification tools from computer science, such as model checking. Motivated by this, the use of transient control methodologies that fulfil the aforementioned qualitative constraints becomes evident. In this talk, we review some of our recent results in applying transient control techniques, and in particular control barrier functions, prescribed performance control and model predictive control, to high level robot task planning under spatiotemporal specifications, treating both the case of a single and a multi-agent system. We further review approaches to task decomposition and consider the case when there are discrepancies between the task and communication graph topologies. The results are supported by relevant experimental validations.
     

  • Schedule: Thursday, December 11 | 8:00-9:00
    Location: Asia I+II+III+IV

    AI-Human Games

    We are at the beginning of an AI revolution, whereby digital tools much more capable than what people could imagine just a few years ago are being developed and are already deployed in almost every sector of the economy and every aspect of our social lives. These AI agents promise to perform many tasks autonomously and much more efficiently than humans, provide complementary information and knowledge to professional decision-makers and become flexible, usable assistants to most humans. AI agents are predicted to be everywhere. But little asked is the key question of how they will interact with humans and even more fundamentally, how they will interact with each other, since the humans that they advise or act on behalf of are locked in myriad social relations.

    In this talk we focus on the decision-making performance of AI agents. We propose new metrics for evaluating the performance of AI agents in sequential decision-making and show that minimizing regret for an AI model (under a single-layer self-attention parametrization) implements a no-regret algorithm, hence offers insightful connections to game-theoretic settings. In the second part of the talk, I will focus on how AI recommendations can be combined with human expertise. I will formulate the human-AI collaboration problem as an incomplete information communication game and highlight what types of miscommunications can arise between AI agents and humans and how this can be fixed.
     

  • Schedule: Friday, December 12 | 8:00-9:00
    Location: Asia I+II+III+IV

    Generalization in Reinforcement Learning: From Foundational Results to New Frontiers

    Reinforcement learning (RL) and optimal control share a deep intellectual heritage, centered on the design and analysis of algorithms for sequential decision-making under uncertainty. This talk will provide a high-level overview of the theoretical foundations of modern RL, focusing on the significant progress that has been made in understanding generalization, sample efficiency, and computational tractability. We will examine foundational results on efficient learning algorithms and their fundamental limits, concluding with a look at new frontiers and posing the question of what role control theory can play in advancing the capabilities of large language models at the heart of modern AI systems.